P wave tomography of the mantle under the Alpine ... and Morelli 2003... · P wave tomography of the mantle under the Alpine-Mediterranean area Claudia Piromallo and Andrea Morelli

P wave tomography of the mantle under the

Alpine-Mediterranean area

Claudia Piromallo and Andrea MorelliIstituto Nazionale di Geofisica e Vulcanologia, Rome, Italy

Received 10 January 2002; revised 5 June 2002; accepted 9 September 2002; published 1 February 2003.

[1] We study the upper mantle P wave velocity structure below the Euro-Mediterraneanarea, down to 1000 km depth, by seismic travel time tomography. We invert summaryresiduals constructed with both regional and teleseismic first arrival data reported by theInternational Seismological Centre (ISC) (1964–1995), introducing some alternativestrategies in the travel time tomographic approach and a new scheme to correct teleseismicdata for global mantle structure. Our high-resolution model PM0.5 is parameterized withthree-dimensional (3-D) linear splines on a grid of nodes with 0.5� spacing in bothhorizontal directions and 50 km vertical spacing. We obtain about 26% root-mean-square(RMS) reduction of residuals by inversion in addition to roughly 31% reduction aftersummary rays formation and selection. Sensitivity analyses are performed through severaltest inversions to explore the resolution characteristics of the model at different spatialscales. The distribution of large-scale fast anomalies suggests that two different stages of aconvection process presently coexist in very close regions. The mantle dynamics ofwestern central Europe is dominated by blockage of subducted slabs at the 660 kmdiscontinuity and ponding of seismically fast material in the transition zone. Contrarily, inthe eastern Mediterranean, fast velocity material sinks into the lower mantle, suggestingthat the flow of the cold downwelling here is not blocked by the 660 km discontinuity. On asmaller scale, the existence of tears in the subducted slab (lithospheric detachment) allalong both margins of the Adriatic plate, as proposed by some authors, is not supported byour tomographic images. INDEX TERMS: 7203 Seismology: Body wave propagation; 7218

Seismology: Lithosphere and upper mantle; 8180 Tectonophysics: Evolution of the Earth: Tomography; 9335

Information Related to Geographic Region: Europe; KEYWORDS: travel time, body wave tomography, upper

mantle, Europe-Mediterranean area, Earth structure

Citation: Piromallo, C., and A. Morelli, P wave tomography of the mantle under the Alpine-Mediterranean area, J. Geophys. Res.,

108(B2), 2065, doi:10.1029/2002JB001757, 2003.

1. Introduction

[2] The Alpine-Mediterranean region developed undercontinuous rearrangement with time of the oceanic spaceduring the progressive shortening of the Tethyan belt. Theclosure of the Tethys ocean started with subduction ofoceanic lithosphere below the southern Eurasian plate atthe time of opening of the central Atlantic (Jurassic–Creta-ceous boundary, about 140 Myr), and continued until thebeginning of Oligocene. On the site of this Mesozoicseaway, several stages of subduction, oceanic accretionand continental collision, originated the main tectonicfeatures of a complex domain. A location map and namesof the main tectonic units are given in Figure 1a. Large-scale reconstructions of the evolution of the Alpine-Medi-terranean region, based on syntheses of kinematic, paleo-magnetic, and geological data, are discussed by Dercourt etal. [1986] and Dewey et al. [1989] and, in spite of the manyunsolved uncertainties, are widely accepted in their main

points. Before Oligocene, though continental collision wasalready present between Apulia and Eurasia, and betweenArabia and Eurasia, most of the convergence was taken upby sinking of oceanic lithosphere along the Eurasian activemargin. The tectonic style east of Apulia was subductiondominated. During Oligocene, oceanic basins remainedprogressively trapped in a wide area, which later on, byEarly Miocene, underwent the passage from subduction-dominated to continental collision-dominated convergence.The orogenic activity in the Alps and the Pyrenees wasparalleled by the development of the European CenozoicRift System (ECRIS) [Ziegler, 1992]. The formation of theApennines began through northwestward subduction ofoceanic lithosphere and eastward migration of the trench.After Oligocene, to the east, subduction progressed belowthe Hellenic-Cyprean Arc and Taurides, possibly initiated inthe Black and southern Caspian Seas, and ceased in theDinarides and Pontides. To the west, by Early Miocene,extension between the Sardinia-Corse Block and Iberiadetermined the formation of new oceanic crust and openingof the Ligurian and Alboran Sea. The Tyrrhenian Seaopened during late Miocene to Pliocene, as a later exten-

JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 108, NO. B2, 2065, doi:10.1029/2002JB001757, 2003

Copyright 2003 by the American Geophysical Union.0148-0227/03/2002JB001757$09.00

ESE 1 - 1

Figure 1. (a) Location map of the Euro-Mediterranean region and nomenclature of the main tectonic unitshere referred to. AnB =Antalya Basin, AlB = Alboran Basin, AM =Armorican Massif, Ar = Arabian Plate,Bal = Balcans, Bet = Betics, BM = Bohemian Massif, BSh = Baltic Shield, CA = Calabrian Arc, CAl =central Alps, CAp = central Apennines, Cau = Caucasus, CM = Massif Central, CyA = Cyprus Arc, DS =Dead Sea, EAl = eastern Alps, ECar = eastern Carpathians, EEP = east European platform, HAt = HighAtlas, HeA = Hellenic Arc, MAt = Middle Atlas, MP = Moesian Platform, NAp = northern Apennines,PaB = Pannonian Basin, PoB = Po Basin, Pon = Pontides, PrB = Provencal Basin, Py = Pyrenees, RhM =RhodopeMassif, Rif = Rif, RM=RhenishMassif, SAp = southern Apennines, SaAt = Saharian Atlas, SC =Sicily Channel, SCar = southern Carpathians, SCB = Sardinia-Corse Block, Tau = Taurides, TeAt = TellAtlas, TuAt = Tunisian Atlas, VrZ = Vrancea Zone, VT = Valencia Trough, WA = western Alps, WCar =western Carpathians, ZZ = Zagros Zone. (b) ISC seismic stations location inside the Euro-Mediterraneanarea. (c) Shallow, h� 50 km, earthquake epicenters from ISC Bulletin data (1964–1995) after relocation insp6 reference model (see text for details). Oblique cylindrical projection centered at 10�W, 45�N with poleat 170�E, 45�N is used.

ESE 1 - 2 PIROMALLO AND MORELLI: P WAVE TOMOGRAPHY OF THE MANTLE

sional episode driven by the retreat of the Calabrian trench[Malinverno and Ryan, 1986]. To date, oceanic space is stillpresent within the collisional belt, both in terms of recentlyformed oceanic crust (Algero-Provencal Basin and Tyrrhe-nian Basin) and of oceanic remnants of Mesozoic age(eastern Mediterranean) [Horvath and Berckhemer, 1982].Subduction of this old lithosphere is still occurring belowthe Calabrian and Hellenic Arcs.[3] Seismic tomography developed in the last two deca-

des to map the three-dimensional (3-D) heterogeneousvelocity structure of the Earth’s interior. Through tomog-raphy on a regional and global scale, seismology is able toprovide useful information to answer questions of geo-dynamics and tectonics. Indeed, tomography yields morerealistic estimates of the amount of subducted lithospherethan seismicity alone can do, detects possible upwellings ofhot (low velocity) material from different depths in themantle, images roots of orogens and allows to argue aboutthe style of convection. The Alpine-Mediterranean area is awide and complex geophysical laboratory where manyquestions regarding these topics are still open. In fact,numerous studies appeared in the literature with the goalof imaging the mantle compressional and shear velocitystructure of the area, from scales ranging from regional tolocal, using different methods and data. On a regional scale,P wave velocity structure has been studied for example byRomanowicz [1980], Hovland et al. [1981], Granet andTrampert [1989], Spakman [1991], Spakman et al. [1993],Piromallo and Morelli [1997], and Bijwaard et al. [1998]. Swave velocity structure was revealed through inversion ofsurface wave phase or group velocities by Panza et al.[1980], Calcagnile and Panza [1990], Ritzwoller and Lev-shin [1998], and Curtis et al. [1998] and through waveforminversion of surface waves by Snieder [1988] and body andsurface waves [Zielhuis and Nolet, 1994; Marquering andSnieder, 1996]. Tomographic studies of the lithosphere–mantle structure of smaller areas exploited data from localnetworks or temporary deployments of arrays: in the Betic-Alboran region [Blanco and Spakman, 1993; Seber et al.,1996a, 1996b; Calvert et al., 2000], in the ECRIS [Glahn etal., 1993; Granet et al., 1995; Ritter et al., 2001], in theItalian peninsula [Panza and Mueller, 1979; Babuska et al.,1990; Amato et al., 1993; Selvaggi and Chiarabba, 1995;Solarino et al., 1996; Lucente et al., 1999], in the Carpa-thian region [Fan et al., 1998; Wenzel et al., 1998], and inthe Hellenic-Aegean area [Ligdas and Main, 1991; Papa-zachos et al., 1995; Papazachos and Nolet, 1997; Tiberi etal., 2000]. Though local studies can obtain an extremelygood resolution, they are however limited in depth and tosmall geographical areas, thus hindering sometimes a com-prehensive image of the deep structure.[4] These studies show that the complex tectonic structure

of the region is reflected on the heterogeneous distribution ofseismic velocities in the lithospheric and sublithosphericmantle. Finer detail is usually revealed by tomographicmodeling of the travel times of P waves, characterized byhigh resolution because of wide data availability and becauseof the short wavelength. Roecker et al. [1993] and Spakmanet al. [1993] suggested the importance of the joint inversionof local/regional and teleseismic travel times to reach bettervertical resolution in upper mantle structure. In a previoustomographic study [Piromallo and Morelli, 1997], using

data from the Bulletins of the International SeismologicalCentre (ISC), we proposed a different, finer, parameter-ization and a different processing scheme with respect tothose adopted until then. Our results yielded a preliminarymodel for 3-D P wave velocity heterogeneities, which willbe referred to as PM97P hereafter. In the present work theinvestigated volume is broadened both in horizontal andvertical dimensions, and the grid of nodes over which themodel is defined has a different orientation to get animproved regional distance ray coverage in the central andeastern part of the model. The model reaches 1000 km depth(300 km deeper than the previous model PM97P), thusimproving the ray sampling, not only at depth. Six yearsof additional data (1990–1995) with respect to PM97P areused. A new scheme to correct teleseismic data for globalmantle structure is introduced.[5] This paper presents the new model, and a discussion

of resolution and robustness of its main features. We firstdescribe the data and the method, then the ability of our datato image test structures at different spatial wavelengths, andwe finally illustrate the tomographic model with horizontalmaps at different depths, and vertical cross sections slicedthrough selected areas. We provide a discussion of ourresults and a comparative analysis both on a large andsmall scale, although a detailed presentation and interpreta-tion of the tomographic model in terms of tectonics andgeodynamics is out of the scope of this paper and will begiven elsewhere.

2. Data Selection and Earthquake Relocation

[6] The convoluted tectonic setting of the Mediterraneanregion originates a complex pattern of seismic activity.Seismicity is characterized by the occurrence of frequentlow to moderate magnitude events, and occasional largemagnitude earthquakes. As illustrated in Figure 1c, the mostactive areas are interplate regions located at the marginsbetween the main interacting plates (Africa, Eurasia, andArabia) and microplates (Anatolia and Iberia) of the Med-iterranean domain. However, intraplate shallow seismicity isalso present. Shallow earthquakes have a rather uniformdistribution over the area (Figure 1c). Deep and intermedi-ate-depth events are instead clustered in the Hellenic Arc(depth �180 km), Cyprus Arc (depth �130 km), CalabrianArc (depth �500 km), Betic-Rif (depth �160 km and threeisolated events at �600 km depth), and eastern Carpathians(depth �220 km). The pattern of both shallow seismicity(Figure 1c) and station distribution in the area (Figure 1b)justifies our choice for the volume to be investigated. Wedefine the cell model with a regular spacing in a coordinatesystem rotated with respect to the geographical one, as givenby an oblique cylindrical projection, centered at (10�W,45�N) with pole at (170�E, 45�N). This choice is intendedto have the model area lying astride the new Equator, whichallows to obtain cells of approximately the same size in bothdirections. The model horizontal dimensions are 6600 km inE-W direction, 3900 km in N-S direction, approximately,and 1000 km in depth. The orientation and aspect ratio of thevolume allow to optimize the sampling by regional rays.[7] P wave travel times reported by the ISC for the time

period 1964–1995 constitute our data set for the presentwork [International Seismological Centre, 1997]. During

PIROMALLO AND MORELLI: P WAVE TOMOGRAPHY OF THE MANTLE ESE 1 - 3

the six additional years with respect to the data set used inPM97P a significant number of new stations have beendeployed throughout the Mediterranean, highly improvingcoverage in some key areas (i.e., Morocco, Tunisia, Syria,and Turkey). This study will benefit by a good ray samplingespecially in the central and western Mediterranean Basin.The extreme paucity, or lack, of stations in the centraleastern part of the North African margin does not insteadallow a satisfactory illumination below the eastern Medi-terranean basins (see Figure 1b).[8] Earthquakes are carefully selected on the basis of a

few simple but stringent criteria, meant to sort out only thewell recorded ones. Each event is required to have thefollowing characteristics: (1) a shallow hypocentral depth(�50 km), (2) at least 30 first arrivals reported by stations atregional or teleseismic distance, and (3) a largest openazimuth of no more than 2 over 8 adjacent azimuthal sectors(in the worst case it can be close to 180�). Application ofthese selection criteria limits location bias due to uneven raypath distribution and ensures more homogeneity in earth-quake geographical distribution and path coverage; inter-mediate-depth and deep earthquakes are rejected becausethey are heavily clustered in small volumes. Focal depth isnot always very reliable in bulletin data [e.g., Engdahl et al.,1998]. However, as we only use shallow hypocenters, this isnot expected to have an appreciable effect on our modelbecause direct P first arrivals are in fact rather insensitive tosource depth (due to trade-off with origin time). For allanalyses, we only retain arrivals in the epicentral distancerange 3� � � � 90� to avoid crustal Pg arrivals and core–mantle boundary diffractions.[9] The ISC Bulletin is the most widely used data set for

travel time tomography at this wide scale, since it includes alarge number of earthquakes and stations, and a massivecollection of arrival times. It is a rather heterogeneous dataset (varied instrumentation and technology for data acquis-ition, different analysts, and picking procedures). Its arrivaltimes are affected by random and systematic sources oferrors and data quality may vary significantly [i.e., Grand,1990; Gudmundsson et al., 1990; Rohm et al., 1999]. Inspite of this, ISC data set is the one with the highest possibleresolution for this kind of studies.[10] We process all selected earthquakes, relocating them

using the radial global Earth model sp6 [Morelli andDziewonski, 1993], to remove the systematic bias in hypo-central parameters due to the ISC location procedure basedon JB travel time tables [Jeffreys and Bullen, 1940]. Thelocation is performed by means of an iterative least squaresoptimization of first arrival residuals. Initial residuals arecalculated using hypocenters and arrival times given by theISC reports, and travel times computed in the sp6 model.Each first arriving phase is always associated to thetheoretical first arrival at that distance. Pn arrivals followthe same association criterion as the other first arrivals,regardless of any identifier reported in the ISC data set.Tables of travel times as a function of slowness areinterpolated using the t-spline method [Buland and Chap-man, 1983] to find the theoretical travel time for thereference model at any given epicentral distance. Sourcedepth is fixed and standard corrections such as for ellipticity[Dziewonski and Gilbert, 1976] and station elevation areapplied.

[11] Worldwide available first arrival P phases, at tele-seismic and regional distance, are used to locate the events.The procedure adjusts origin time and epicentral coordi-nates to minimize the summed squared weighted residualsfor each event, following the uniform reduction scheme[Jeffreys, 1932], with average distribution parametersappropriate for regional and teleseismic P residuals [Pir-omallo and Morelli, 1998, 2001]. Event mislocations can becomputed with respect to a selected set of ground-truth testevents [Piromallo and Morelli, 2001]. Root-mean-square(RMS) difference between known hypocentral parametersand those found after 1-D location in sp6 model is 16.39km for earthquakes and 9.95 km for explosions recorded atteleseismic distance. RMS values of this magnitude, whichare an improvement with respect to the bulletin estimates,represent the largest advancement that can be attained usingjust a 1-D reference Earth model. The effects of relocationin sp6 and in a heterogeneous Earth are specificallyaddressed by Piromallo and Morelli [1998, 2001].[12] We locate over 52,000 selected events using more

than 5,000,000 observations. All rays having either sourceor receiver in the study area, are then stored for subsequentuse.

3. Tomographic Method

3.1. Summary Rays and Static Corrections

[13] Summary residuals are often used, rather than indi-vidual observations, as input data to the inversion becausesummary rays reduce spatial imbalance in data coverage,they minimize noise related to small-scale crustal hetero-geneities, and they limit the redundancy of data in theinversion, at the same time retaining the geographicallydependent character of travel times [i.e., Morelli andDziewonski, 1991; Robertson and Woodhouse, 1995]. Tocompute summary residuals the Earth’s surface is subdi-vided in cells and all individual rays connecting the samepair of cells, regardless of which of the two contains thesource and which the receiver, are grouped into a summaryray. To this purpose, we use two nested grids: 1656approximately equi-areal trapezoidal cells on the wholeglobe (cell size is 5� � 5� at the Equator), and 8400smaller cells (0.5� � 0.5�) in the Mediterranean area.Summary rays therefore stem from each cell of the targetarea containing stations, events, or both. This strategyincreases the coverage with respect to studies based onlyon rays travelling to stations inside the area. To eachsummary ray a residual is then associated, defined as theweighted average of all individual residuals contained in thebundle [Piromallo and Morelli, 2001], where the dataweighting function is the uniform reduction weight speci-fied by Jeffreys [1932].[14] Residuals of regional distance summary rays (� �

28�) are used in the inversion only if they travel entirelywithin the volume, in order to prevent mapping of lateralheterogeneities close, but external, to the inversion domaininto our model. Moreover, only summary rays with residualless than 3 s, and composed by at least 3 individualobservations, are retained as input to the inversion. Thisprocedure and further selection isolate a total of 112,139(52,514 regional and 59,625 teleseismic) summary rays, outof 334,723. The RMS reduction of residuals due to sum-


mary ray formation before inversion is about 31%, theinitial RMS being 1.44 s.[15] Due to the large-scale character of this study, ray

paths of teleseismic observations do not entirely lay withinthe model volume, thus providing a sampling of both thetarget volume and of the mantle heterogeneities locatedoutside. An erroneous mapping of these anomalies into ourmodel may occur, unless we apply an appropriate correctionfor teleseismic data. To this purpose, regional tomographicstudies often use station static corrections, either as modelparameters in a joint inversion for the 3-D velocity structure[e.g., Spakman et al., 1993] or as independent parameters ina stepwise sequence [e.g., Calvert et al., 2000]. Otherstudies reduce the effect of outside structure by embeddingtheir regional model in a global (more coarsely parameter-ized) mesh [Fukao et al., 1992; Widiyantoro and van derHilst, 1996] or by doing global inversions with an adaptivegrid to achieve high resolution locally [Bijwaard et al.,1998; Sambridge and Gudmunsson, 1998; Karason and vander Hilst, 2001].[16] We follow a different approach and devise a scheme

which allows the use of teleseismic rays, by introducing astraightforward alternative to a global inversion to correct forstructure outside the model volume. For each 5� � 5� cell ofour global grid, we estimate the contribution of the externalmantle structure to the total delay, by averaging all summaryresiduals from summary rays connecting that cell to anysmall cell in the Alpine-Mediterranean region. Our mantlecorrections represent a path-integrated account of large-scalevelocity structure, as a result of a selective sampling of themantle in the ‘‘fat beam’’ based on the 5� � 5� cell, on oneend, and the 35�� 60� base of themodel volume, on the otherend. This is better than using static teleseismic stationcorrections, which instead give a global sampling of thewhole mantle as seen from each station. In addition, ourreciprocal summary rays, combining sources and stations in

each cell, permit the calculation for several more cells atteleseismic distance.[17] Mantle corrections are plotted in Figure 2. Analo-

gously to station statics [e.g., Robertson and Woodhouse,1997] our selective mantle corrections represent a firstapproximation account of large-scale mantle velocity struc-ture. Each correction, shown in Figure 2, is the result of aselective sampling of the mantle by paths travelling from,or to, the Mediterranean, and therefore represents thecontributions of long-wavelength lower mantle structure,and shorter-wavelength (�5�) upper mantle structure out-side the model volume. Some general features of thesemantle corrections are nonetheless similar to classic stationstatics [Dziewonski and Anderson, 1983; Hager and Clay-ton, 1989; Zhou and Wang, 1994] insofar they are corre-lated with lithospheric structure beneath the remotereceiver or cell: negative delays therefore mark shieldareas, and positive delays are present in younger tectoni-cally active continental regions, and across ocean ridgesand rifting areas. The mantle corrections range from �2.9to +4.2 s, with RMS 0.64 s and average 0.24 s. Thispositive regional average is a consequence of the uppermantle beneath the Mediterranean being slower than theglobal reference, and has no effect for our goal ofreconstructing lateral variations of velocity, but should betaken into account if absolute velocities are sought. Forscarcely populated cells, up to about 10 summary rays, thenorm of regionally averaged cell terms decreases as thenumber of summary rays increases [Piromallo and Mor-elli, 2001], but this effect is not crucial for more populatedcells.

3.2. Parameterization and Inversion

[18] We parameterize the perturbation of the slownessfield by means of local basis functions, represented by 3-Dlinear splines on a 3-D lattice of 180,411 nodes (121 and

Figure 2. Mantle corrections computed for roughly equal area 5� � 5� cells used to account foraspherical velocity structure outside the model volume. Pluses and circles indicate positive and negativecorrections, respectively. Each correction is obtained through a selective sampling of the mantle by pathstravelling to the Mediterranean from a cell located at teleseismic distance outside the investigated region(see text for details).


71 nodes, respectively, in the EW and NS direction,21 nodes in depth). The horizontal spacing is again 0.5�in both horizontal directions, while the vertical spacing is50 km. In general, the use of a continuous interpolatorgives a more realistic description of the velocity field, ifcompared to a block-wise model with the same number ofparameters [Nolet, 1996]. Travel times of P waves, withperiod of the order of 1 s and wavelength of about 10 kmin the mantle, are able to resolve structures of 30–80 kmwavelength at upper mantle depths and of 100–200 km atlower mantle depths [Nolet, 2000]. Our parameterization,with node interspacing of about 55 km, is thus potentiallysuitable for describing features not smaller than this size.Of course, structures are not necessarily resolved to theblock size (nominal resolution), as will be explained inthe next section. Moreover, the station and event distri-bution in the Alpine-Mediterranean region (some of the0.5� � 0.5� squares may not even contain any seismo-graphic station or earthquake), may map shorter-wave-length structure into a larger-scale, smoother model, andhence introduce some artifacts in the upper layers incorrespondence of these scarcely sampled areas. However,a fine grid is the best measure we can take to limit thisaliasing effect.[19] The forward problem is solved by 1-D ray tracing

through a ray shooting algorithm in spherical geometry insp6 background model. Pn ray paths result analytically asthe effect of refraction at the 35 km deep Moho. In the areaencompassed by this study, velocity anomalies rarelyexceed a few percent, so that 1-D ray tracing can beconsidered a legitimate approximation. The use of a 1-Dray tracing algorithm may have a smoothing effect on theresulting model of a strongly heterogeneous region of theupper mantle. A rough estimate of this smoothing effect isgiven by Bijwaard and Spakman [2000, Figure 3a]. Differ-ences in RMS amplitudes between models resulting after 3-D or 1-D ray tracing are of the order of 0.2% in the crustallayer (35 km) and attain a maximum 0.5% in the litho-spheric layers (70,120 km), below which they becomenegligible.[20] The solution to the inverse problem is sought by

least squares minimization through the iterative LSQRalgorithm [Paige and Saunders, 1982; Nolet, 1987].Though LSQR is intrinsically damped, we introduce anadditional explicit roughness damping through an isotropicgradient minimization condition, to constrain ill-deter-mined model parameters. We assume that observationalerrors are Gaussian and uncorrelated, with standard devia-tion 1 s, and we assign an a priori probability densitydistribution for the change of seismic velocity betweenany couple of adjacent nodes shaped as a Gaussian withzero mean and standard deviation 0.5%. The gradientconstraint is applied both horizontally and vertically; onlythe top layer (at 0 km depth) is kept unconstrained, inorder to limit the contamination of the layer below (at 50km depth) by unmodeled crustal heterogeneities. Theoverall regularization applied is thus a combination ofnorm and gradient damping, isotropic and without anydepth dependence.[21] Our P velocity model PM0.5 is obtained after 15

iterations of the LSQR algorithm. The RMS reduction inthe residuals due to data fit in the inversion is about

26%, comparable to 28% obtained by Spakman et al.[1993].

4. Sensitivity Analyses

[22] Ray paths provide a rather inhomogeneous illumina-tion of the mantle structure. Before inspecting the results ofthe inversion, it is therefore important to analyze the pathdistribution over the model volume. Figure 3 depicts, inmap view and logarithmic scale, the ray density for somelayers of the model. Ray density is the cumulative value ofthe partial derivative of slowness with respect to travel time,a quantity analogous to the cell hit-count in studies based onblock parameterization.[23] The consequence of the inhomogeneous distribution

of seismicity and stations in this area on ray illumination isclearly seen from a comparison of Figure 3 with Figures 1band 1c. There is substantial difference, especially in theshallower layers, in the coverage provided by teleseismicrays, confined to areas right below stations and events, andregional rays, more extensively sampling the regions inbetween the ray foci. In fact, the 50 km depth layer islargely dominated by horizontally travelling Pn rays, sam-pling the lithosphere and crossed by vertical teleseismic raysin correspondence of stations and events (which give thespot-like features in some regions). Earthquakes in theHellenic-Aegean region, recorded by European stations,provide an extremely good coverage of the central westernpart of the model, although most rays have a SE-NW trend.At 150 km, the ray coverage is more uniform, but themaximum density is lower than above. In both layers, thecoverage of the African margin, the eastern European plat-form, and the Arabian Plate is limited to small areas,unevenly sampled, due to sparse station and event locationsand to the location of these regions in the periphery of thedomain. At intermediate depths (250 km) the illuminationgaps in these regions are partially filled by incoming tele-seismic rays, and we observe a more uniform ray density allover the area. This is even more evident moving from 350to 450 km depth, with the fairly illuminated area extendingnorthward and eastward. Moving to deeper layers (600–700 km) ray sampling gradually extends to the border of theregion, practically achieving a full coverage of the area byteleseismic rays in the lowermost 200 km. An enhancedsampling in the layer below each discontinuity (450 and700 km) can be noticed, due to the refraction at theinterfaces of regional rays and subsequent subhorizontalsampling.[24] In the case of a manageable number of parameters,

the best way to assess the reliability of the estimated modelis to compute the full resolution matrix, which contains thewhole information, and the a posteriori error matrix. Whenwe are faced with an very large number of unknowns,presenting a readable matrix is however a problem. We canalternatively resort to ‘‘sensitivity tests’’ [e.g., Spakman andNolet, 1988; van der Hilst and Engdahl, 1991; van der Hilstet al., 1993; Leveque et al., 1993], which address theresolution of the inversion scheme in terms of its abilityto retrieve a known input model, given the same raycoverage which will be used for the real data inversion.Sensitivity analyses allow for a comprehensive, relativelyrapid, visual inspection in the spatial resolution of a model


with a large number of parameters, provided that theassumptions made hold true. However, they only providequalitative estimates of accuracy and resolution, and theyonly show a limited representation of the resolution matrix.

The synthetic travel times used to retrieve a sample structureare computed with the same theoretical apparatus, andsimplifications, that are used for the actual inversion, anddo not therefore test the effect of ‘‘errors’’ in the theory.

Figure 3. Ray density (cumulative value of the partial derivative of slowness with respect to travel time)distribution in a logarithmic scale. The ray density field does not vary rapidly with depth. Therefore, weonly display maps located at representative depths: two shallow layers (50 and 150 km), an intermediateone (250 km), one layer above and one below each upper mantle seismic discontinuity (350, 450, 600,and 700 km), and one in proximity of the base of the model (900 km). Solid black is used for regions ofno sampling. See text for comments.


Furthermore, only specific features (e.g., chessboard andharmonic fields) can be used, and therefore only sensitivityto those particular shapes, very different from the real Earthstructure, can really be addressed [Leveque et al., 1993]. Inspite of these shortcomings, sensitivity tests have become astandard way of addressing model resolution as their resultscan be promptly visualized.[25] We perform several synthetic tests, in order to study

the resolution for patterns of long, intermediate and shortwavelengths of geodynamical interest and for sharp orsmooth anomaly contrasts. The input synthetic models areconstructed as regular patterns of velocity variations relativeto the surrounding reference velocity profile. Travel timesare then computed by forward calculation through thesynthetic model, and Gaussian random noise, with standarddeviation s, is added to the data before inversion. The valueof s for the Gaussian noise is chosen to approximatelyreproduce the signal-to-noise ratio of the actual data, asexplained in detail by Piromallo and Morelli [1998]. Thesame inversion parameters as for actual data are thenapplied, i.e., number of iterations and regularization. Theirvalues are reported in Table 1, together with the variancereduction obtained in the inversion for each test and for thereal data.[26] We show here the results of chessboard tests where

input heterogeneities of alternating sign (±4%) are super-imposed on the ambient velocity profile. The positive/negative anomaly is attributed to the same number of nodesin each direction, separated by an equal number of zeroamplitude anomaly nodes. Gaussian random noise, withstandard deviation s = 0.85 s, is added to the data. Differentwavelength patterns are tested, among which we show tworepresentative results. The first test, in the left panels ofFigure 4, is aimed at estimating retrieval of small-scalestructures (boxcar anomalies of two nodes in each direction,approximately 110 km in size). Horizontal smearing ofanomalies of this size occurs in regions where ray samplingis poor and/or strongly anisotropic. All over the depth rangeof the model this effect is particularly pronounced in theRed Sea, Arabian Plate, and eastern European platform,sometimes even joining the anomalies over several nodes.In the Northern African margin, the western and easternMediterranean, northern Europe, Black Sea, eastern Turkey,Caucasus and Zagros region smearing is still present but

less critical. In the remaining regions, small-scale hetero-geneities are satisfactorily retrieved in the depth rangesampled by both regional and teleseismic rays, while theiramplitude decays in the lowermost 200 km of the model.Vertical leaking of anomalies is confined to the closest layerin areas of good sampling and at shallow depths. In eachlayer, amplitudes of the anomalies are strongly dependenton the ray density, as may be expected. In the right panels ofFigure 4, larger-scale structures (four-node anomalies,approximately 220 km, in each direction) are analyzed.With respect to the previous pattern, this is more easilyretrievable in amplitude, and the area over which reliableimages are obtained is enlarged horizontally and extends tothe bottom of the model (though some noise distorts theshape of the input model). For example also the westernMediterranean, Northern African margin, Black Sea, Tur-key, Caucasus, and Zagros region are nicely retrieved.Moving to the deeper layers, the area of recovered featuresextends to the north, NE, but the portion of Africa andArabian plate still remains unresolved or highly smeared.[27] We also did a harmonic model recovery test. The input

pattern is given by 2-D smoothly varying features, obtainedby superposition of sine functions, with amplitudes rangingbetween ±4% and delay times are again contaminated byGaussian random noise before inversion (see Table 1).Wavelengths of �300 and �600 km in both horizontaldirections are considered. Figure 5 shows the RMS ampli-tude of both retrieved models (lengths �300 and �600 km)and of the input model, as a function of depth. The RMSprofiles indicate that retrieved amplitudes are rather uniformwith respect to depth, the only exceptions being at the topand bottom of the model. The amplitude of the layer at50 km depth is better recovered, due to the strong samplingprovided by Pn rays. The two extremal layers, at 0 and1000 km, have instead lower amplitudes, a geometric effectdue to the lower sampling provided by rays to the nodeslocated at the boundaries of the inversion volume (only theray segments inside the volume are considered). The layerat 0 km is also affected by a more irregular sampling, due toray clustering at stations and hypocentral areas. The RMSamplitudes of the reconstructed model are about 60% of theinput pattern, larger for broader anomalies.

5. Main Features of Model PM0.5

[28] For sake of completeness, and to preserve the 3-Dcontinuity of its features, the resulting model is displayed inits totality, by means of map views at all the different depthlevels. Model PM0.5 is expressed as percentage velocityperturbation with respect to the reference model sp6. Theaverage anomaly over a horizontal layer is always sub-tracted for clarity. Layer averages are shown in Figure 6 as afunction of depth. Since the values are always rather small,less than 0.2%, the new average P velocity profile is almostundistinguishable from the background model sp6.[29] An overall decreasing structural complexity and

amplitude weakening is observed moving from the surfaceto depth, since the Earth structure is particularly heteroge-neous with strong regional variations in the uppermost 250km. This is clearly illustrated by Figure 7a, where we showthe RMS amplitude of model PM0.5 and of separate testinversions of teleseismic and regional data, plotted as a

Table 1. Parameters and Characteristics of the LSQR Inversion of

Real Data and Synthetic Modelsa

Model

Input Anomaly

s (s)

RMS of Residuals

Length(km)

Amplitude(%)

Before(s)

After(s)

Reduction(%)

Real data – – – 0.99 0.73 26chess 1 �110 4 0.85 0.85 0.72 14chess 2 �200 4 0.85 0.85 0.72 14harmo 1 �300 4 0.85 0.99 0.72 27harmo 2 �600 4 0.85 1.18 0.72 39impulse �110 7 0.40 0.63 0.52 17

aThe size and amplitude of the input velocity anomalies for the sensitivitytests are specified together with the standard deviation of the Gaussiannoise (s). The damping parameter and the number of iterations in theinversion are always the same as for the data inversion. The RMS values ofresiduals before and after inversion are listed together with the RMSreduction in percent, both for the real data and for the synthetic models.


Figure 4. Chessboard recovery test in map views at some selected depths. Input heterogeneities ofopposite sign (whose location is indicated by ±2% contouring) are alternately superimposed on theambient velocity profile and are marked by contouring. The input anomaly is attributed to the samenumber of nodes (2 nodes in the left panels, 4 nodes in the right ones) in each direction, separated by anequal number of unperturbed nodes. Gaussian random noise is added to the data before inversion. Theoutput model is given in gray shades. The rectangle in the top left panel identifies the region best resolvedat all scale lengths.


function of depth. Comparison with Figure 5, showinginstead that input patterns are retrieved with rather uniformRMS amplitudes, indicates that the behavior of Figure 7a isnot an artifact of ray sampling. Overall, model RMSamplitudes decrease rapidly going from the shallow partof the model to larger depth [see also Fukao et al., 2001].The transition zone is marked by a relative increase of RMSamplitudes, mainly retrieved by regional rays. Althoughcharacterized by smaller-depth sensitivity, teleseismic raysdetect the decrease of RMS amplitude with depth, andobviously contribute the most to constrain the model atlarger depths. In Figure 7b we show instead the relativecontribution of positive and negative velocity anomalies tothe RMS amplitude of PM0.5. At shallow depths (�250km) slow anomalies have slightly larger RMS amplitudesthan fast ones. Below 250 km, instead, fast anomalies havelarger RMS amplitudes than slow ones, sensibly higher inthe transition zone, less pronounced in the shallowest lowermantle. High RMS amplitude in the transition zone are onlypresent in positive anomalies.[30] Each map in Figure 8 reports relative P velocity

variations on the horizontal layer of the 3-D grid at thatdepth. Some selected vertical cross sections, obtained fromthe 3-D model with average velocity profile subtracted, are

shown in Figure 9, along with the ray density distributionand an impulse recovery test (two-node input anomalies by±7% in each direction, separated by six unperturbed nodes,see Table 1). Additional material can be found on the Website (http://www.ingv.it/seismoglo/ptomo).[31] The ‘‘crustal’’ layer at 0 km depth is shown but it is

not interpreted since it suffers from the strongly inhomoge-neous distribution of stations and events, the effects ofunmodeled shallow structure, and of smearing from theclose layer at 50 km. The contribution of regional rays isdominant, amplitudes are very large in well sampled areasand velocity contrasts are quite sharp. This velocity patternis affected by a trade-off between variations in crustalthickness and velocity anomalies. Since no crustal correc-tion is taken into account, neglected heterogeneities incrustal thickness (with horizontal scales comparable to thoseof main tectonic features) may induce quite large variationsin the travel times of shallow travelling rays. Smaller-scaleheterogeneities (less than about 50 km) are instead filteredout by the use of summary rays.

5.1. Large-Scale Features

[32] The most prominent large-scale feature of PM0.5 atshallow depth is the difference in the upper mantle structurebetween the Precambrian Baltic and east European Shields,marked by high velocities, and the low velocities of thewestern and central Euro-Mediterranean young tectonicprovinces. The high velocity below the east Europeanplatform and Baltic shield is continuous in PM0.5 from50 to 350–400 km, and particularly pronounced in thedepth range 100–300 km. The western MediterraneanBasin and Europe are instead overall dominated by slowvelocity anomalies between 100 and 400 km depth on alarge scale. This pattern of large-scale shallow anomalies isin accord with results from other P and S tomographicmodels [Romanowicz, 1980; Snieder, 1988; Spakman et al.,

Figure 5. RMS amplitude as a function of depth forharmonic sensitivity tests. The input pattern of the harmonicmodels is given by the superposition, in each layer, of sinefunctions in the two orthogonal directions, giving a 2-Dharmonic slowness anomaly pattern, with amplitudesranging ±4% and characteristic lengths of �300 and�600 km. Delay times are contaminated with Gaussianrandom noise (Table 1). The retrieved harmonic models foranomalies of �300 km (harmo 1) and �600 km (harmo 2)size are shown together with the RMS amplitude of theinput model (dashed line). RMS amplitudes are computedwithin the best resolved region of the model shown inFigure 4 (top left panel).

Figure 6. Averages of velocity anomalies over horizontallayers as a function of depth. These values are subtracted,layer by layer, from the model to produce the map views inFigure 8.


1993; Zielhuis and Nolet, 1994; Marquering and Snieder,1996; Curtis et al., 1998]. However, in our model the sharpgradient of the boundary between east and central Europe,running along the ancient suture of the Tournquist-Teisseyrezone, extends to depths of at least 300 km, larger than inprevious models [Zielhuis and Nolet, 1994].[33] Another dominant feature is the approximately con-

tinuous S-shaped fast anomaly that can be followed, throughthe four layers at 250–450 km depth, from the North Africanmargin, to the Calabrian Arc, the Apennines, the Alps andthe Carpathians. It represents the signature of subductedlithosphere along the Alpine collisional belt [Wortel andSpakman, 1992; Spakman et al., 1993; Piromallo andMorelli, 1997]. In the 500–700 km depth range the raysampling is clearly more uniform for the whole area, due tothe contribution of teleseismic rays. The pattern of fastanomalies of the previous layers joins into a wide and stronghigh velocity feature, approximately located in the centralpart of the model. The most remarkable heterogeneity liesbeneath the area interested by the Alpine orogeny, theTurkish promontory and, to the east, the Zagros Zone. Alarge amount of fast velocity material stagnant in thetransition zone below Europe [e.g., Spakman et al., 1993;Marquering and Snieder, 1996] is particularly evident inPM0.5 (see also cross section Cc, Figure 9), and could be theremnant of the Tethyan oceanic lithosphere.[34] Only teleseismic rays illuminate the structure below

the 670 km discontinuity. At these depths teleseismic raysare not as vertical as in the lithosphere, therefore we do notexpect too strong vertical smearing. Two deep large-scaleanomalies are present in the 850–1000 km depth range: theslow velocity region under northwestern Europe and thehigh velocity area from the Ionian Sea to Turkey. The lowvelocity is part of a broader and deeper slow anomalypossibly related to a lower mantle upwelling [Goes et al.,1999]; the fast velocity anomaly corresponds instead to thewesternmost edge of the Tethyan subduction belt [Grand et

al., 1997; van der Hilst and Karason, 1999; Faccenna etal., 2003]. The bottom layers of the model are the ones mostprone to contamination from neighboring, external, Earthstructure. The good agreement of our results with whole-mantle models, however, supports the robustness of thesefeatures. We regard to this result as to a significant verifi-cation of the effectiveness of mantle travel time corrections,and of the reliability of both approaches.

5.2. Small-Scale Features

[35] On a smaller scale, fast velocity anomalies character-ize the lithosphere of the basins. Some of them are wellconfined to the shallowest layer at 50 km (like the westernMediterranean, Adriatic and Ionian Basin), some otherspropagate to the layer below at 100 km (as Black andCaspian Sea), or even down to 150–200 km (as in theeastern Mediterranean Basin, where however some verticalsmearing is likely). Low velocities at these depths markinstead areas of rifting (as the ECRIS, the Sicily Channel,and the Dead Sea), back arc regions (as the Alboran, theTyrrhenian and the Aegean Sea, and the Pannonian Basin)and magmatic provinces (as the volcanic provinces of centralItaly and central Europe). A fast velocity feature, laterallydiscontinuous, along the North African margin is imaged,though at the border of the illuminated region, gainingamplitude from shallow (50 km) to deeper layers (400 km).[36] Beneath the Alboran region, seismically slow mate-

rial at shallow depth (50–150 km) is underlain by a singlecoherent fast body, SW-NE oriented, apparently continuousfrom 250 down to 700 km depth. Most of the recent larger-scale seismic tomography studies agree in detecting a largepositive anomaly interpreted as the image of a subductedslab [Spakman et al., 1993; Blanco and Spakman, 1993;Piromallo and Morelli, 1997; Bijwaard et al., 1998]. Ourmodel also reveals significant anomalies in northern Africa,in correspondence of the Rif region, in the 50–150 kmdepth range (see Figure 8), previously overlooked by larger-

Figure 7. (a) RMS amplitude of model PM0.5 velocity anomalies and of the separated inversions forteleseismic and regional rays only, plotted as a function of depth. (b) RMS amplitude of PM0.5 velocitymodel, as a function of depth, computed separately for positive (pluses) and negative (dashed line)anomalies. RMS amplitudes are computed within the best resolved domain marked in the left top panel ofFigure 4.


Figure 8. Map views of our best 3-D model PM0.5. P velocity perturbation with respect to reference model sp6 isdisplayed. Not illuminated areas are simply blanked out. The bottom layer of the model at 1000 km depth is not shown,being very similar to the shallower one (950 km). See text for details.

ESE

1-12

PIROMALLO

AND

MORELLI:PWAVETOMOGRAPHY

OFTHEMANTLE

Figure 8. (continued)

PIROMALLO

AND


OFTHEMANTLE

ESE

1-13


ESE

1-14

PIROMALLO

AND


OFTHEMANTLE


PIROMALLO

AND


OFTHEMANTLE

ESE

1-15


ESE

1-16

PIROMALLO

AND


OFTHEMANTLE

Figure 9. Cross sections through some of the main features of the velocity model. The top mapillustrates the section location. Each cross section has the same length (approximately 20� along a greatcircle) and its vertical extent reaches the bottom of the model. Ticks along the line segment indicate thedistance from the extremities by 5� intervals. Below the map, three panels are plotted: the top panel forthe model, the central panel for the impulse recovery test results (contour levels of the input model at±1% and ±3% are shown), and the bottom panel for the ray density distribution. Ticks at 5� intervalsalong the upper axis of each panel correspond to the locations along the section line segment plotted onthe map. The 410 and 660 km discontinuities are indicated as a thin and thick lines, respectively. (Aa)Low velocity heterogeneity beneath the Massif Central (topmost 250 km). The synthetic test shows thatthe central portion of the section is well resolved, but some vertical leaking occurs and the narrow,conduit-like structure is probably not a robust feature. (Bb) Section running approximately along theEuropean Geotraverse [Blundell et al., 1992]. A fast feature deepens from the surface below the centralAlps, its strongest portion reaching about 200 km depth. From ray sampling and the impulse recoverytest, we expect a good resolution in the central portion of the section. Notice also the accumulation of fastmaterial between 400 and 800 km. (Cc) Shallow fast anomaly below the northern Adriatic Sea possiblyconnected with the fast structure, below the northern Apennines. On the Tyrrhenian side, a wedge-shapedslow anomaly just above the slab merges with the lower amplitude western Mediterranean slow anomalydown to 400 km depth. (Dd) Strong slow anomaly beneath the central Apennines (0–250 km depth). Tothe east, the fast Adriatic lithosphere deepens below the Dinarides. (Ee) High velocity anomaly on theIonian side of the Calabrian Arc connected to the fast structure steeply dipping below the arc into themantle and bending horizontally in the transition zone. The synthetic test indicates that horizontalsmearing may result in artifacts along this section at midmantle depths. (Ff) Fast material can be followedfrom the surface through the upper mantle, across the 670 km discontinuity and down to the bottom of themodel in the Hellenic Trench. A large amplitude, well sampled, low anomaly wedge is found incorrespondence of the Aegean Basin. Correlation between density of rays and shape of input anomaliesindicates that smearing of these small anomalies likely occurs in the deeper part.


scale models. Seber et al. [1996a] image a high velocitybody beneath the Alboran Sea extending to the eastern Rifarea as well, from subcrustal depths down to 350 km(bottom of their model). However, the limited verticaldimension of their model does not allow to evaluate theeffective depth extent of the fast anomaly, which we showto enter into the transition zone. Calvert et al. [2000],though imaging fast anomalies also in the deep layers oftheir model (down to 670 km), discard the existence of awhole coherent body, favoring instead the hypothesis of twosmaller pieces of delaminated lithospheric material.[37] Pronounced low velocity anomalies are displayed by

PM0.5 in the shallow layers below the Massif Central,Rhenish Massif and Bohemian Massif, coalescing into abroad anomaly in the 150–250 km depth range. The bulk ofthis low velocity perturbation resides in the shallowest200–250 km and gradually attenuates in amplitude whileapproaching the transition zone. These heterogeneities incentral Europe cover a wide region in correspondence of theECRIS volcanic province, a belt of extensional thinning andrifting [Wilson and Downes, 1992; Ziegler, 1992]. Belowthe French Massif Central (cross section Aa, Figure 9) themain low velocity body extends from the surface down to

250 km depth, showing very good agreement with the localhigh-resolution teleseismic model of Granet et al. [1995]. Inthe Eifel region (Rhenish Massif ), instead, Ritter et al.[2001] find evidence of low velocities down to at least400 km, deeper than in PM0.5. Some studies, combiningtomographic images with signatures from isotope geochem-istry of mafic volcanic rocks of this province, led toreconsider active mantle upwellings as sources of riftingand volcanism [Granet et al., 1995; Hoernle et al., 1995;Goes et al., 1999], as suggested in previous literature[Duncan et al., 1972; Froidevaux et al., 1974]. However,the detection of a possible direct, lower mantle, origin ofsuch upwellings by means of seismic studies is hampered,on one side, by the insufficient resolution at depth of larger-scale tomographic models, and, on the other hand, by thelimited depth extent of higher-resolution local studies.Although below the Massif Central and the Rhine regionwell-resolved slow anomalies are also present at the base ofPM0.5 (see map views at 700–1000 km), the velocitypattern is rather weak and heterogeneous in the transitionzone below the ECRIS. Robust evidence for a direct lowermantle feeding (i.e., a conduit-like feature) of the shallowlow velocity anomalies is missing. Where a weak stem-like



feature is imaged, such as below the Massif Central, it is atthe very limit of our resolution.[38] The whole Alpine chain is underlain by high veloc-

ities in a continuous belt down to 150 km. In the westernAlps a sharp boundary is detected between slow velocityalong the northwestern border of the chain, and a strong fastanomaly to the east. In the deeper layers (200–250 km)anomalies are weaker and the western and eastern Alpsappear to be horizontally disconnected from each other andfrom the fast anomaly below the Dinaric chain. A detailedmapping of these shallow anomalies is absent in the resultsby Spakman et al. [1993], likely due to the coarser param-eterization of their model. Fast velocity features rejoin atdepth (starting at about 300 km) and connect to the wholeAlpine fast belt. In the western and central Alps thesignature of the fast anomaly is particularly pronouncedfrom 50 down to 300–350 km (cross section Bb, Figure 9)and is unanimously interpreted as the ‘‘root’’ of the Alps,the Eurasian lithosphere southwardly subducted below theAdriatic promontory [Kissling and Spakman, 1996]. Acrossthe eastern Alps instead we do not find evidence of such aclear pattern of subduction: the high velocity features of theEurasian and Adriatic lithospheres both seem to be onlyweakly connected with the deep seated (100–400 km) body.The difference between the deep structure of western andeastern Alps [Panza and Mueller, 1979; Babuska et al.,1990] is here imaged with better detail than in studies basedon teleseismic data alone [Amato et al., 1993; Solarino etal., 1996; Lucente et al., 1999]. Moreover, according to ourtests, imaging capabilities of PM0.5 should not be cruciallydifferent between the western and eastern arcs.[39] Along the Apenninic chain two major velocity fea-

tures are recognized: while in the northern part a high velocitystructure is particularly evident in the layers between 100 and200 km (cross section Cc, Figure 9), in the southern part aslow anomaly is instead found at these depths (cross sectionDd, Figure 9). To the south, right below the Calabrian Arc, afast velocity structure can be seen again in the shallow layers(100–200 km depth). Cross section Ee (Figure 9) shows thevertical extent and continuity of this fast feature, lyinghorizontal in the Ionian domain, steeply dipping to the NWin the mantle and then bending to almost horizontal in thetransition zone. At 250 km and deeper, northern and southernApennines join into the single, fast anomaly belt runningsouthward with continuity through the Calabrian and NorthAfrican margin, and northward through the Alps and Carpa-thians. A pronounced shallow slow anomaly is detected onthe Tyrrhenian side, in correspondence of the back arc region,interpreted as the evidence of an asthenospheric wedge [Meleet al., 1998;Hearn, 1999]. While in the Calabrian portion thesubducted lithosphere is imaged with continuity by the vastmajority of tomographic models [Amato et al., 1993; Spak-man et al., 1993; Selvaggi and Chiarabba, 1995; Piromalloand Morelli, 1997; Lucente et al., 1999], vertical continuityof the slab in its shallow portion below the Apenninic belt isinstead a point of discrepancy. Since some models [Spakman,1991; Spakman et al., 1993] show no apparent connectionbetween the Adriatic plate and the subducted lithosphereunder the Apennines, Wortel and Spakman [1992] proposeda model of lateral migration of slab detachment from north-western Italy to the Calabrian Arc. However, we note that thesame northern Apenninic cross section shows enhanced

vertical continuity in model BSE [Bijwaard et al., 1998]with respect to EUR89B [Spakman et al., 1993] (see thestudy of Carminati et al. [1998, Figure 7c]), displaying avelocity structure much closer to the one retrieved in bothPM97P and PM0.5. Also teleseimic studies do not revealany significant gap in the subducted lithosphere below thecentral Apennines [Amato et al., 1993; Lucente et al., 1999];however, in these models the signal of a shallow positiveanomaly could be easily obscured by smearing along tele-seismic ray paths of a deeper, consistent, fast anomaly.[40] The entire Carpathian chain is underlain by slow

anomalies at shallow depths of PM0.5 (50 km), in sharpcontrast with the fast velocities of the cratonic east Euro-pean platform. In the eastern Carpathians, below the Vran-cea zone, a weak fast velocity anomaly is present at 100 kmdepth, becoming more pronounced in the following layer(150 km). As depth increases, positive velocities tend todominate below the whole chain, as a hook-shaped anomalyin map views (200–250 km). Regional tomographic studieshave focused their attention on the Vrancea zone, alsoimaging a fast body in correspondence of the intermedi-ate-depth earthquake foci, interpreted as a piece of sub-ducted lithosphere [Oncescu et al., 1984; Fan et al., 1998;Wenzel et al., 1998]. However, lack of resolution does notallow to establish the vertical extent of the high velocitybody in these models. Only larger-scale studies like PM0.5allow to detect anomalies at depth larger than the seismi-cally active part of the slab, and they suggest that the deepportion of the slab may easily extend down to 300 km.Images of the shallow velocity structure retrieved by differ-ent models of the Vrancea zone led to different interpreta-tions of the origin and nature of the fast velocity anomaly:the remnant of a slab of oceanic lithosphere now detachedfrom the surface [Wortel and Spakman, 1992; Wenzel et al.,1998], or instead a small fragment of oceanic lithospherestill attached to a larger piece of continental lithosphere[Fan et al., 1998]. Wortel and Spakman [2000] favor theview of a continuous slab down to 350 km. Our model doesnot show, instead, any robust continuity to the surface of theVrancea fast anomaly. This is an interesting open question,which needs further attention.[41] The Hellenic Arc is marked by positive anomalies,

approximately following its changes in strike. In the shallowlayers these heterogeneities show continuity to the NW withthe fast velocities below the Dinaric chain. The highvelocity anomaly can be easily followed from shallowlithospheric depths across the transition zone and below,to the bottom of the model (cross section Ff, Figure 9),especially in correspondence of the W-SW portion of theHellenic Arc. The high velocity anomaly sinks at an angleof about 45� in the upper mantle, while its shallower portionappears to dip on a less steep plane. As depth increases, fastmaterial along the arc converges towards the northernAegean (map views) and appears to be only partly deflectedin correspondence of the 660 km discontinuity (crosssection Ff, Figure 9). At 50 km depth the whole AegeanSea in the back arc setting is dominated by a wide,pronounced, slow anomaly, which shifts northward andattenuates at increasing depths, down to about 250 km.The Cyprus Arc is well delimited at shallow depths by thecold Levantine lithosphere. A small fast velocity anomaly,NW of Cyprus, joins across the Antalya Basin, in corre-


spondence of the trench’s cusp, with the more pronouncedanomaly below Turkey, which can be followed down to thetransition zone. The fast anomaly below the Hellenic Arc,interpreted as African lithosphere subducted beneath theAegean, is imaged by both regional studies [Spakman et al.,1988; Granet and Trampert, 1989; Ligdas and Main, 1991;Spakman, 1991; Piromallo and Morelli, 1997; Tiberi et al.,2000] and larger-scale models, which show that it reachesthe lower mantle [Spakman et al., 1993; van der Hilst et al.,1997; Bijwaard et al., 1998; this work]. Since the observa-tion by Spakman et al. [1988] of a discontinuity in thesubducted slab at a depth of 100–250 km, as a horizontaltear in the north and western portions of the arc, thehypothesis of detachment of the deeper part of the slabfrom the surface has been put forward [Wortel and Spak-man, 1992, 2000]. However, from the map views of ourmodel we see a possible interruption in the fast anomalyonly at 100–150 km in a very limited area below mainlandGreece and it is hard to assess its effective robustness.Moreover, high-resolution local studies of the crust andtopmost mantle [Papazachos et al., 1995; Papazachos andNolet, 1997] do not detect any gap at all in the slab at theseshallow depths.

6. Discussion

[42] The presence of seismically fast material in thetransition zone and below raises the question about thenature of such heterogeneities, and on the style of present-day mantle convection in the Alpine-Mediterranean region.Lateral variations in seismic velocities reported at all depthsin the mantle from tomographic observations have beenusually interpreted in terms of thermal structure related toconvection. Though both temperature and composition mayinfluence wave speeds, most of the 3-D mantle velocityvariation can probably be attributed to changes in temper-ature [e.g., Ranalli, 1996], the effect of mantle compositioncontributing only less than 1% [Sobolev et al., 1997; Goeset al., 2000]. Interpreting the deep high velocity anomaliesas mainly related to ‘‘cold’’ material, the snapshot of thepresent-day large-scale features given by tomographic pic-tures can provide important information both in the frame-work of global geodynamics and in the discussion ofspecific evolutionary scenarios for the Mediterranean. Thepattern of large-scale anomalies revealed by PM0.5 sug-gests that two different convection stages presently coexistin very close regions. The mantle dynamics of westerncentral Europe are dominated by layering, with blockage ofsubducted slabs at the 660 km discontinuity and consequentaccumulation of seismically fast material in the transitionzone [Piromallo et al., 2001]. On the contrary, in theeastern Mediterranean a substantial amount of fast velocitymaterial appears to sink into the lower mantle (even deeperthan 1000 km, according to van der Hilst et al. [1997]),suggesting that the flow of the cold downwelling here is notbarred by the 660 km discontinuity. The extent of mantlelayering under Europe has been estimated through waveletanalysis of the seismic anomalies of PM0.5 by focussing onthe different length scales in the correlation function [Pir-omallo et al., 2001]. Between the depths of 500 and 600 kmunder western central Europe this correlation analysisshows that a strong correlation for long length scales, of

around 400 km, exists over a wide area of about 2000 �4000 km, while weaker correlation characterizes shorterlength scales (150 km). On the other hand, between layersat depths of 400 and 600 km the correlation deteriorates onthe long length scales and becomes even worse at the shortlength scales. These results suggest that the thickness of therecumbent fast (cold) material in the transition zone isbetween 100 and 150 km. Moreover, the correlationbetween the layer at 600 km and those at larger depthshows a clear shift of the strong coherence to the easternMediterranean, where indeed a substantial amount of fastvelocity material appears to sink into the lower mantle [seealso van der Hilst et al., 1997]. A possible explanation ofthis scenario is that in the eastern Mediterranean, the longerduration of subduction (active from the Cretaceous orpossibly the Jurassic rather than from the Tertiary) andthe different motion of the upper plate–trench system withrespect to the western Mediterranean, favored the sinking inthe mantle of a large amount of lithospheric material and itspiling up at the upper–lower mantle boundary. Accumu-lation would overcome the effects of the viscosity jump andof the endothermic phase transition, causing the eventualflush of material to the lower mantle. The past subductionhistory which possibly led to the present-day scenario andthe implications for the style of mantle convection areinspected in detail elsewhere [Faccenna et al., 2003], tyingtogether geological data, paleotectonic reconstructions,plate motion and tomographic data.[43] In some tomographic models [Spakman, 1990; Spak-

man et al., 1993] the absence of continuity in the images ofstructures identified as subducted slabs has been interpretedby the authors in the common frame of a near surfacelithospheric detachment process, characterizing and influ-encing the tectonic evolution of the Euro-Mediterraneanregion [Wortel and Spakman, 1992]. However, in manycases the interruption in the distribution of the high velocityanomalies does not appear to be a consistent feature amongdifferent studies. More recent, better resolved, models seemto exhibit more vertical continuity of fast structures thanolder models. Many causes may affect the continuity of slabimages: inhomogeneous sampling, data density, referencemodel, parameterization, damping, ray tracing technique.As shown by sensitivity analyses vertical gaps in tomo-graphic images of slabs as large as 50–60 km or less cannotbe quantitatively resolved by regional-scale tomographicmodels [Spakman, 1990; Spakman et al., 1993; Piromalloand Morelli, 1997; Bijwaard et al., 1998; Lucente et al.,1999; this work]. Though of course the existence of smallfeatures cannot be ruled out, too detailed interpretation oftomographic models, which may go beyond the resolution,should be avoided. Sticking to robust features, model PM0.5shows a strong, robust, difference between the northern andsouthern Apennines in the top 200 km. While below thenorthern portion of the chain we find a fast anomaly at alldepths, below the southern part a pronounced, wide andwell-resolved slow anomaly is present. Along the Dinaric-Hellenic chain, instead, a pronounced fast velocity anomalycan be followed at all depths in the central southern portion,while a slow anomaly dominates the northernmost part downto the top of the transition zone. Therefore, the hypothesis oflateral migration of slab detachment, proposed byWortel andSpakman [1992, 2000] as a unifying process taking place on


both sides of the Adriatic plate, is not supported by ourresults, which depict instead a more complex scenerydeserving further investigation.

7. Concluding Remarks

[44] Mantle structure in the Euro-Mediterranean regionhas been studied at different levels of detail by manyauthors for more than two decades. This study was moti-vated by the goal of further enhancing the detail of thetomographic image, in a wide-scale model. While confirm-ing the general view of the lithosphere and mantle structureof the region, we also present important differences thatmay provide new insight into active geodynamic processes.[45] The main 3-D large-scale and small-scale features

detected by our model PM0.5 are generally in fair corre-spondence with those obtained by other P wave studies withsimilar or slightly lower resolution. Fair agreement is foundwith model EUR89B [Spakman et al., 1993], spanningapproximately the same area as PM0.5. However, thecoarser parameterization of EUR89B and the referencemodel adopted are responsible for some apparent differ-ences in the detailed structure. Spakman et al. [1993] used abackground reference model with a low velocity layer in the120–250 km depth range and a small negative velocitygradient. Fast velocity anomalies at sublithospheric depthsare often more discontinuous in EUR89B than in studiesbased on background models with a positive velocitygradient throughout the upper mantle. A low velocity layeris not present in modern global reference models [Kennettand Engdahl, 1991; Morelli and Dziewonski, 1993; Kennettet al., 1995], nor we find evidence for it in the averagevelocity profile for our region. In fact, the average radialvelocity perturbation (Figure 6) is so small that the newprofile does not differ appreciably from the reference model.Enhanced vertical continuity is shared, for instance, by BSE[Bijwaard et al., 1998] and PM0.5.[46] We have shown that there are good similarities

between our model and BSE [Bijwaard et al., 1998], resultof a global inversion with variable resolution. A maindifference between models BSE and PM0.5, on one side,and the class of previous models, on the other side, consistsof explicit allowance, in newer approaches, of heterogene-ous structure in the whole mantle. BSE is obtained throughactual inversion for global P velocity structure; whereas inPM0.5 we instead focus on inversion in a specific geo-graphic region, but we minimize the effect of global Earthstructure outside the study volume (be it heterogeneity oranisotropy or a systematic station bias) through the use ofaveraged path corrections. Both approaches have advantagesand disadvantages, and can be seen as aimed at partlydiffering goals. A global-scale inversion represents theconceptually most complete approach, and today it can bepushed to quite high resolution by modern availability ofcomputing power, but it has to face an increased complexityand to tackle with potential sources of disturbance such asthe extremely irregular data coverage of the globe. Ourapproach, although of course limited to the study of aspecific geographical region with good data coverage, hasthe advantage of extreme computational simplicity andbypasses sources of instabilities, at the cost of a less faithful,ad hoc, description of external Earth structure. BSE differs

from PM0.5 also for many other choices, among which themost important are the EHB data set [Engdahl et al., 1998],the parameterization in irregular blocks, the depth extent ofthe model (2900 km), the nominal resolution in the uppermantle (up to 0.6�), the reference model adopted (ak135), thejoint inversion for velocity, relocation vectors and stationcorrections, the explicit regularization applied (depth andcell volume dependent second derivative damping). With somany procedural differences, agreement of the results hasgood significance.[47] The linearized approach of our tomographic method

is another simplification we adopt. In presence of sharpvelocity gradients as in the shallowest portion of the Earth,data cannot be considered insensitive to approximations ofray geometry and 3-D ray tracing may be required to obtainmore accurate images. The nonlinear effect is proved to beparticularly significant for strongly heterogeneous models,producing an overall focusing of structures: a slight changein anomaly patterns, a small increase in variance reductionand model amplitudes, and a baseline shift towards lowvelocities [Papazachos and Nolet, 1997; Bijwaard andSpakman, 2000]. However, resulting images do not differconceptually from the outcome of a linearized inversionwith 1-D ray tracing [Bijwaard and Spakman, 2000]. Theuse of regional phases, which propagate in very heteroge-neous structures, where velocity perturbations may bestronger than a few percent, implies a probable smoothingeffect, but no important mistakes, induced by the 1-Dalgorithm. The interpretation of regional seismic phases isproblematic also because their onset is often difficult to pickdue, for example, to triplication. Alternative approaches,avoiding this difficulty, involve fitting entire waveforms orsurface wave phase and group velocity observation, butcannot reach the spatial resolution attained by travel timetomography.[48] Amplitudes of velocity anomalies are not very well

constrained and sensitivity test results suggest that RMSamplitudes of the reconstructed model may be as small as60% of the input pattern. Moreover, they decrease as raysampling decreases, an effect inherent to the inversionalgorithm. The retrieved perturbation may underestimatethe actual values because of the effect of regularization(damping) of the inversion, trade-off with earthquake loca-tion (the hypocenter parameters during the preliminarylocation can in principle have absorbed substantial signal)and the neglect of ray-bending effects, which provides adefocusing effect. The inability to assess with confidencethe amplitudes of velocity anomalies is one of the seriouslimits of tomographic imaging, still hindering more quanti-tative interpretation of tomographic results.

[49] Acknowledgments. We appreciate the critical review by Robvan der Hilst. This paper benefited from comments by Michel Granet andLapo Boschi, which greatly improved the manuscript. We acknowledgefruitful discussions with Wim Spakman, Claudio Faccenna, and Dave Yuen.We thank Enzo Boschi for his continuing encouragement and support. Allfigures were created using Generic Mapping Tools (GMT) [Wessel andSmith, 1991].

ReferencesAmato, A., B. Alessandrini, and G. B. Cimini, Teleseismic tomography ofItaly, in Seismic Tomography: Theory and Practice, edited by H. M. Iyerand K. Hirahara, pp. 361–396, Chapman and Hall, New York, 1993.


Babuska, V., J. Plomerova, and M. Granet, The deep lithosphere in the Alps:A model inferred from P residuals, Tectonophysics, 176, 137–165, 1990.

Bijwaard, H., and W. Spakman, Non-linear global P-wave tomography byiterated linearized inversion, Geophys. J. Int., 141, 71–82, 2000.

Bijwaard, H., W. Spakman, and R. Engdahl, Closing the gap betweenregional and global travel time tomography, J. Geophys. Res., 103,30,055–30,078, 1998.

Blanco, M. J., and W. Spakman, The P-wave velocity structure of themantle below the Iberian Peninsula: Evidence for subducted lithospherebelow southern Spain, Tectonophysics, 221, 13–34, 1993.

Blundell, D., R. Freeman, and S. Mueller, A continent revealed, in A Con-tinent Revealed, edited by D. Blundell et al., p. 275, Cambridge Univ.Press, New York, 1992.

Buland, R., and C. H. Chapman, The computation of seismic travel times,Bull. Seismol. Soc. Am., 73, 1271–1302, 1983.

Calcagnile, G., and G. F. Panza, Crustal and upper mantle structure of theMediterranean area derived from surface-wave data, Phys. Earth Planet.Inter., 60, 163–168, 1990.

Calvert, A., E. Sandvol, D. Seber, M. Barazangi, S. Roecker, T. Mourabit,F. Vidal, G. Alguacil, and N. Jabour, Geodynamic evolution of the litho-sphere and upper mantle beneath the Alboran region of the westernMediterranean: Constraints from travel time tomography, J. Geophys.Res., 105, 10,871–10,898, 2000.

Carminati, E., M. J. R. Wortel, W. Spakman, and R. Sabadini, The role ofslab detachment processes in the opening of the western-central Medi-terranean basins: Some geological and geophysical evidence, Earth Pla-net. Sci. Lett., 160, 651–665, 1998.

Curtis, A., J. Trampert, R. Snieder, and B. Dost, Eurasian fundamentalmode surface wave phase velocities and their relationship with tectonicstructures, J. Geophys. Res., 103, 26,919–26,947, 1998.

Dercourt, J., et al., Geological evolution of the from the Atlantic to thePamirs since the Lias, Tectonophysics, 123, 241–315, 1986.

Dewey, J. F., M. L. Helman, E. Turco, D. H. Hutton, and S. D. Knott,Kinematics of the western Mediterranean, in Alpine Tectonics, Geol. Soc.Spec. Publ., edited by M. P. Coward et al., 45, 265–283, 1989.

Duncan, R. A., N. Petersen, and R. Hargraves, Mantle plumes, movementof the European Plate, and polar wandering, Nature, 239, 82–86, 1972.

Dziewonski, A. M., and D. L. Anderson, Travel times and station correc-tions for P waves at teleseismic distances, J. Geophys. Res., 88, 3295–3314, 1983.

Dziewonski, A. M., and F. Gilbert, The effect of small aspherical perturba-tions on travel times and a re-examination of the corrections for ellipti-city, Geophys. J. R. Astron. Soc., 44, 7–16, 1976.

Engdahl, R., R. van der Hilst, and R. Buland, Global teleseismic earthquakerelocation with improved travel times and procedures for depth determi-nation, Bull. Seismol. Soc. Am., 88, 722–743, 1998.

Faccenna, C., L. Jolivet, C. Piromallo, and A. Morelli, Subduction and thedepth of convection in the Mediterranean mantle, J. Geophys. Res., 108,doi:10.1029/2001JB001690, in press, 2003.

Fan, G., T. C. Wallace, and D. Zhao, Tomographic imaging of deep velocitystructure beneath the eastern and southern Carpathians, Romania: Impli-cations for continental collision, J. Geophys. Res., 103, 2705–2723,1998.

Froidevaux, C., R. Brousse, and H. Bellon, Hot spot in France?, Nature,248, 749–751, 1974.

Fukao, Y., M. Obayashi, and H. Inoue, Subducting slabs stagnant in themantle transition zone, J. Geophys. Res., 97, 4822–4909, 1992.

Fukao, Y., S. Widiyantoro, and M. Obayashi, Stagnant slabs in the upperand lower mantle transition region, Rev. Geophys., 39, 291–323, 2001.

Glahn, A., and M. Granet, Southern Rhine Graben: Small-wavelength to-mographic study and implications for the dynamic evolution of the gra-ben, Geophys. J. Int., 113, 399–418, 1993.

Goes, S., W. Spakman, and H. Bijwaard, A lower mantle source for centralEuropean volcanism, Science, 286, 1928–1931, 1999.

Goes, S., R. Govers, and P. Vacher, Shallow mantle temperatures underEurope from P and S wave tomography, J. Geophys. Res., 105, 11,153–11,169, 2000.

Grand, S., A possible station bias in travel time measurements reported toISC, Geophys. Res. Lett., 17, 17–20, 1990.

Grand, S. P., R. D. van der Hilst, and S. Widiyantoro, High resolutionglobal tomography: A snapshot of convection in the Earth, GSA Today,7, 1–7, 1997.

Granet, M., and J. Trampert, Large-scale P velocity structures in the Euro-Mediterranean area, Geophys. J. Int., 99, 583–594, 1989.

Granet, M., M. Wilson, and U. Achauer, Imaging a mantle plume beneaththe Massif Central (France), Earth Planet. Sci. Lett., 136, 281–296,1995.

Gudmundsson, O., J. H. Davies, and R. W. Clayton, Stochastic analysis ofglobal traveltime data: Mantle heterogeneity and random errors in the ISCdata, Geophys. J. Int., 102, 25–43, 1990.

Hager, B. H., and R. W. Clayton, Constraints on the structure of mantleconvection using seismic observations, flow models, and the geoid, inMantle Convection, edited by W. R. Peltier, pp. 657–763, Gordon andBreach, Newark, N. J., 1989.

Hearn, T. M., Uppermost mantle velocities and anisotropy beneath Europe,J. Geophys. Res., 104, 15,123–15,139, 1999.

Hoernle, K., Y.-S. Zhang, and D. Graham, Seismic and geochemical evi-dence for large-scale mantle upwelling beneath the eastern Atlantic andwestern and central Europe, Nature, 374, 34–39, 1995.

Horvath, F., and H. Berckhemer, Mediterranean back-arc basins, in Alpine–Mediterranean Geodynamics, Geodyn. Ser., vol. 7, edited by H. Berkhe-mer and K. J. Hsu, pp. 141–174, AGU, Washington, D. C., 1982.

Hovland, J., D. Gubbins, and E. S. Husebye, Upper mantle heterogene-ities beneath Central Europe, Geophys. J. R. Astron. Soc., 66, 261–284,1981.

International Seismological Centre, Bulletin Disks 1–6, Int. Seismol. Cent.,Thatcham, UK, 1997.

Jeffreys, H., An alternative to the rejection of observations, Proc. R. Soc.London, Ser. A, 187, 78–87, 1932.

Jeffreys, H., and K. E. Bullen, Seismological Tables, Br. Assoc. for the Adv.of Sci., London, 1940.

Karason, H., and R. D. van der Hilst, Tomographic imaging of the lower-most mantle with differential times of refracted and diffracted core phases(PKP, Pdiff ), J. Geophys. Res., 106, 6569–6588, 2001.

Kennett, B. L. N., and E. R. Engdahl, Traveltimes for global earthquakelocation and phase identification, Geophys. J. Int., 105, 429–465,1991.

Kennett, B. L. N., E. R. Engdahl, and R. Buland, Constraints on seismicvelocities in the Earth from traveltimes, Geophys. J. Int., 122, 108–124,1995.

Kissling, E., and W. Spakman, Interpretation of tomographic images ofuppermost mantle structure: Examples from the western and central Alps,J. Geodyn., 21, 97–111, 1996.

Leveque, J.-J., L. Rivera, and G. Wittlinger, On the use of checker-boardtest to assess the resolution of tomographic inversions, Geophys. J. Int.,115, 313–318, 1993.

Ligdas, C. N., and I. G. Main, On the resolving power of tomographicimages in the Aegean area, Geophys. J. Int., 107, 197–203, 1991.

Lucente, F. P., C. Chiarabba, G. B. Cimini, and D. Giardini, Tomographicconstraints on the geodynamic evolution of the Italian region, J. Geophys.Res., 104, 20,307–20,327, 1999.

Malinverno, A., and W. B. F. Ryan, Extension in the Tyrrhenian sea andshortening in the Apennines as result of arc migration driven by sinkingof the lithosphere, Tectonics, 5, 227–245, 1986.

Marquering, H., and R. Snieder, Shear-wave velocity structure beneathEurope, the northeastern Atlantic and western Asia from waveform in-versions including surface-wave mode coupling, Geophys. J. Int., 127,283–304, 1996.

Mele, G., A. Rovelli, D. Seber, T. M. Hearn, and M. Barazangi, Compres-sional velocity structure and anisotropy in the uppermost mantle beneathItaly and surrounding regions, J. Geophys. Res., 103, 12,529–12,543,1998.

Morelli, A., and A. Dziewonski, Joint determination of lateral heterogeneityand earthquake location, in Glacial Isostasy, Sea-Level and MantleRheology, edited by R. Sabadini et al., pp. 515–534, Kluwer Acad.,Norwell, Mass., 1991.

Morelli, A., and A. Dziewonski, Body wave traveltimes and a sphericallysymmetric P- and S-wave velocity model,Geophys. J. Int., 112, 178–194,1993.

Nolet, G., Seismic wave propagation and seismic tomography, in SeismicTomography, edited by G. Nolet, pp. 1–23, D. Reidel, Norwell, Mass.,1987.

Nolet, G., A general view on the seismic inverse problem, in SeismicModelling of the Earth Structure, edited by E. Boschi et al., pp. 1–27,Ed. Compositori, Bologna, Italy, 1996.

Nolet, G., Interpreting seismic waveforms: Forward and inverse problemsfor heterogeneous media, in Problems in Geophysics for the Next Millen-nium, edited by E. Boschi et al., pp. 373–401, Ed. Compositori, Bologna,Italy, 2000.

Oncescu, M. C., V. Burlacu, M. Angel, and V. Smalbergher, Three-dimen-sional P-wave velocity image under the Carpathian arc, Tectonophysics,106, 305–319, 1984.

Paige, C. C., and M. A. Saunders, LSQR: An algorithm for sparse linearequations and sparse least squares, ACM Trans. Math. Software, 8, 43–71,1982.

Panza, G. F., and S. Mueller, The plate boundary between Eurasia andAfrica in the Alpine Area, Mem. Sci. Geol., 33, 43–50, 1979.

Panza, G. F., S. Mueller, and G. Calcagnile, The gross features of thelithosphere–asthenosphere system in Europe from seismic surface wavesand body waves, Pure Appl. Geophys., 118, 1209–1213, 1980.


Papazachos, C., and G. Nolet, P and S deep velocity structure of theHellenic area obtained by robust nonlinear inversion of travel times,J. Geophys. Res., 102, 8349–8367, 1997.

Papazachos, C. B., P. M. Hatzidimitriou, D. G. Panagiotopoulos, and G. N.Tsokas, Tomography of the crust and upper mantle in southeast Europe,J. Geophys. Res., 100, 12,405–12,442, 1995.

Piromallo, C., and A. Morelli, Imaging the Mediterranean upper mantle byP-wave travel time tomography, Ann. Geofis., XL, 963–979, 1997.

Piromallo, C., and A. Morelli, P-wave propagation heterogeneity and earth-quake location in the Mediterranean region, Geophys. J. Int., 135, 232–254, 1998.

Piromallo, C., and A. Morelli, Improving earthquake location: An alterna-tive to three-dimensional structural models, Pure Appl. Geophys., 158,319–347, 2001.

Piromallo, C., A. Vincent, D. Yuen, and A. Morelli, Dynamics of thetransition zone under Europe inferred from wavelet cross-spectra of seis-mic tomography, Phys. Earth Planet. Inter., 125, 125–139, 2001.

Ranalli, G., Seismic tomography and mineral physics, in Seismic Modellingof the Earth Structure, edited by E. Boschi et al., pp. 443–459, Ed.Compositori, Bologna, Italy, 1996.

Ritter, J. R. R., M. Jordan, U. R. Christensen, and U. Achauer, A mantleplume below the Eifel volcanic fields, Germany, Earth Planet. Sci. Lett.,186, 7–14, 2001.

Ritzwoller, M. H., and A. L. Levshin, Eurasian surface wave tomography:Group velocities, J. Geophys. Res., 103, 4839–4878, 1998.

Robertson, G. S., and J. H. Woodhouse, Evidence for proportionality ofP and S heterogeneity in the lower mantle, Geophys. J. Int., 123, 85–116, 1995.

Robertson, G. S., and J. H. Woodhouse, Comparison of P and S stationcorrections and their relationship to upper mantle structure, J. Geophys.Res., 102, 27,355–27,366, 1997.

Roecker, S. W., T. M. Sabitova, L. P. Vinnik, Y. A. Burmakov, M. I.Golvanov, R. Mamatkanova, and L. Munirova, Three-dimensional elas-tic-wave velocity structure of the western and central Tien Shan, J. Geo-phys. Res., 98, 15,779–15,795, 1993.

Rohm, A. H. E., J. Trampert, and R. K. Snieder, Bias in reported seismicarrival times deduced from the ISC, Geophys. J. Int., 137, 163–174,1999.

Romanowicz, B. A., A study of large-scale lateral variations of P-velocityin the upper mantle beneath western Europe, Geophys. J. R. Astron. Soc.,63, 217–232, 1980.

Sambridge, M., and O. Gudmunsson, Tomography with irregular cells,J. Geophys. Res., 103, 773–781, 1998.

Seber, D., M. Barazangi, A. Ibenbrahim, and A. Demnati, Geophysicalevidence for lithospheric delamination beneath the Alboran Sea and theBetic-Rif mountains, Nature, 379, 785–790, 1996a.

Seber, D., M. Barazangi, B. A. Tadili, M. Ramdani, A. Ibenbrahim, andD. Ben Sari, Three-dimensional upper mantle structure beneath theintraplate Atlas and interplate Rif mountains of Morocco, J. Geophys.Res., 101, 3125–3138, 1996b.

Selvaggi, G., and C. Chiarabba, Seismicity and P-velocity image of thesouthern Tyrrhenian subduction zone, Geophys. J. Int., 121, 818–826,1995.

Snieder, R., Large scale waveform inversions of surface waves for lateralheterogeneity, 2, Application to surface waves in Europe and the Medi-terranean, J. Geophys. Res., 93, 12,067–12,080, 1988.

Sobolev, S. V., H. Zeyen, M. Granet, U. Achauer, C. Bauer, F. Werling,R. Altherr, and K. Fuchs, Upper mantle temperatures and lithosphere–asthenosphere system beneath the French Massif Central constrained by

seismic, gravity, petrologic and thermal observations, Tectonophysics,275, 143–164, 1997.

Solarino, S., D. Spallarossa, S. Parolai, M. Cattaneo, and C. Eva, Litho-asthenospheric structures of northern Italy as inferred from teleseismicP-wave tomography, Tectonophysics, 260, 271–289, 1996.

Spakman, W., Tomographic images of the upper mantle below centralEurope and the Mediterranean, Terra Nova, 2, 542–553, 1990.

Spakman, W., Delay-time tomography of the upper mantle below Europe,the Mediterranean and Asia Minor, Geophys. J. Int., 107, 309–332,1991.

Spakman, W., and G. Nolet, Imaging algorithms, accuracy and resolution indelay time tomography, in Mathematical Geophysics, edited by N. J.Vlaar et al., pp. 155–187, D. Reidel, Norwell, Mass., 1988.

Spakman, W., M. J. R. Wortel, and N. J. Vlaar, The Hellenic subductionzone: A tomographic image and its geodynamic implications, Geophys.Res. Lett., 15, 60–63, 1988.

Spakman, W., S. van der Lee, and R. van der Hilst, Travel-time tomographyof the European-Mediterranean mantle down to 1400 km, Phys. EarthPlanet. Inter., 79, 3–74, 1993.

Tiberi, C., et al., Crustal and upper mantle structure beneath the Corinth rift(Greece) from a teleseismic tomography study, J. Geophys. Res., 105,28,159–28,171, 2000.

van der Hilst, R. D., and E. R. Engdahl, On ISC pP and PP data and theiruse in delay time tomography of the Caribbean region, Geophys. J. Int.,106, 169–188, 1991.

van der Hilst, R. D., and H. Karason, Compositional heterogeneity in thebottom 1000 kilometers of Earth’s mantle: Toward a hybrid convectionmodel, Science, 283, 1885–1888, 1999.

van der Hilst, R. D., E. R. Engdahl, and W. Spakman, Tomographic inver-sion of P and pP data for aspherical mantle structure below the northwestPacific region, Geophys. J. Int., 115, 264–302, 1993.

van der Hilst, R. D., S. Widiyantoro, and R. Engdahl, Evidence for deepmantle circulation from global tomography, Nature, 386, 578–584, 1997.

Wenzel, F., U. Achauer, D. Enescu, E. Kissling, R. Russo, V. Mocanu, andG. Musacchio, Detailed look at final stage of break-off is target of studyin Romania, Eos Trans. AGU, 48, 589–594, 1998.

Wessel, P., and W. H. F. Smith, Free software helps map and display data,Eos Trans. AGU, 72, 441, 1991.

Widiyantoro, S., and R. D. van der Hilst, Structure and evolution of sub-ducted lithosphere beneath the Sunda arc, Indonesia, Science, 271,1566–1570, 1996.

Wilson, M., and H. Downes, Mafic alkaline magmatism associated with theEuropean Cenozoic rift system, Tectonophysics, 208, 173–182, 1992.

Wortel, M. J. R., and W. Spakman, Structure and dynamics of subductedlithosphere in the Mediterranean region, Proc. K. Ned. Akad. Wet., 95,325–347, 1992.

Wortel, M. J. R., and W. Spakman, Subduction and slab detachment in theMediterranean-Carpathian region, Science, 290, 1910–1917, 2000.

Zhou, H., and H. Wang, A revisit to P wave travel time statics at teleseismicstations, J. Geophys. Res., 99, 17,849–17,862, 1994.

Ziegler, P. A., European Cenozoic rift system, Tectonophysics, 208, 91–111, 1992.

Zielhuis, A., and G. Nolet, Shear-wave velocity variations in the uppermantle beneath central Europe, Geophys. J. Int., 117, 695–715, 1994.

��A. Morelli and C. Piromallo, Istituto Nazionale di Geofisica e

Vulcanologia, Via di Vigna Murata 605, I-00143, Rome, Italy. ( [email protected])


Documents

P wave tomography of the mantle under the Alpine ... and Morelli 2003... · P wave tomography of the mantle under the Alpine-Mediterranean area Claudia Piromallo and Andrea Morelli